How aiWare approach to functional safety design reduces system cost
aiWare achieves ASIL B compliance with only 15% silicon area overhead and less than 1% performance impact.
In the rapidly evolving landscape of automated driving (AD), safety isn’t just a priority — it’s a prerequisite. As our industry continues to push boundaries, ensuring functional safety becomes increasingly difficult, with hierarchical systems and varied requirements. aiMotive’s aiWare is the only ASIL B-certified Neural Processing Unit (NPU) Hardware IP for automotive AI — designed ground-up for functional safety compliance.
Understanding the hierarchical nature of safety
In the world of automated driving systems, safety is not a uniform and fixed concept; it varies with each level of autonomy. From the SAE Level 2 to Level 3 and beyond, the safety requirements vary, presenting a multifaceted challenge for engineers and developers.
Example safety requirements for different levels of automated driving
Although an NPU sits relatively low in the overall system hierarchy, it is a key building block for modern automated driving systems that depend on AI for core system functionality — hence, the degree to which it can be trusted to function as expected is vitally important.
Different layers and hierarchy in an automated driving system
With aiWare, we’ve embraced this complexity, ensuring that our solution is agile enough to adapt seamlessly to different levels of safety integrity.
The ground-up approach to safety
aiWare is an automotive-first design, unlike other offerings in the market that retarget automotive use cases as an afterthought. Hence the design doesn’t suffer from various legacy constraints that typically impact other solutions. These constraints often mean that functional safety is retrofitted on top of legacy design, which results in a sub-optimal solution that increases the silicon cost (e.g., using fully duplicated functional units), and/or does not provide sufficient hardware architectural metrics coverage, typically relying on additional mechanisms (e.g., externally supplied BIST or software checks) to make up for the absence of dedicated, hardware-based safety mechanisms (with higher achievable diagnostic coverage). Unlike most conventional NPU solutions that resort to retrofitting functional safety onto their existing designs, with aiWare we took a different path.
Non-ideal approches using retrofitted functional safety design
Purpuse built, optimized functional safety design
We engineered aiWare from the ground up with functional safety as a core principle. This ground-up design allows us to ensure that the safety mechanisms are efficiently designed to minimize our customers’ overheads on system costs without compromising on the requirements they need for their automated driving systems.
aiWare safety certified software and hardware components
Many hardware IPs provide only partial functional safety measures, leaving end customers to invest considerable engineering effort to achieve final compliance. At aiMotive, we’re committed to easing this burden by handling the bulk of the functional safety engineering work ourselves. Our goal is to deliver a fully compliant solution that customers can seamlessly integrate without the need for extensive additional investment.
With that in mind, we ensured that every component of aiWare, from its hardware architecture to its software algorithms, is designed and engineered to fulfill the requirements of ISO26262 for ASIL B. This holistic approach ensures that functional safety isn’t just an afterthought — it’s woven into the very fabric of aiWare’s DNA.
Efficiency without compromise
There is another aspect to the big picture: in the quest for achieving ASIL B, high execution efficiency is often sacrificed in order to achieve compliance. This poses a significant challenge for customers. While some legacy solutions may sacrifice performance for safety or vice versa, with aiWare we wanted to strike the perfect balance between the two.
aiWare’s ground-up design not only prioritizes functional safety but also maximizes efficiency without compromising on either front. By streamlining safety mechanisms and minimizing overhead costs, aiWare enhances the performance of automated driving systems while reducing the burden on customers during their certification process. The end result: aiWare achieves ASIL B compliance with no more than 15% silicon area overhead and less than 1% performance impact.
Conclusion
aiWare isn’t just a repackaged legacy neural processing IP; it is purpose-built to run your AD software efficiently without compromising full ASIL B compliance. We understand the hierarchical and diverse nature of automotive functional safety, embrace a ground-up approach, and prioritize efficiency without compromise. By investing in functional safety, not just substantial cost savings but performance gains can be achieved from day one. In the automotive industry, functional safety must be built in from the start. Otherwise, one will have to pay later with higher engineering costs, timescale delays, and lower-than-expected performance.
Having the industry’s first ASIL B-certified NPU Hardware IP, aiWare, is a testament to our automotive first commitment.
Written by:
Mustafa Ali
aiWare Product Director